5.2 PID Modes, Gain, Reset and Rate

Key Takeaways

  • Gain is the ratio of normalized output change to normalized error change in the proportional term. Proportional band is its inverse expressed on a percent span basis.

  • A P-only loop with fixed bias can settle with offset if the equilibrium command differs from the bias.

  • Reset time is the reciprocal of repeat rate in the dependent form. The parallel integral coefficient also includes controller gain when converting from that form.

  • Anti-windup and tracking depend on the effective applied limits and on how the specific controller is configured.

  • Derivative on PV avoids direct setpoint kick in the derivative term. Measurement noise, filtering and timing still matter.

Last updated: November 2026

PID Control Modes and Controller Settings

ModeInput to the termEffect during operation
ProportionalPresent errorA change in error changes the calculated output in proportion to gain.
IntegralAccumulated errorA constant nonzero error keeps changing this contribution while integration is free to proceed.
DerivativeRate of change of error or measured PV, as configuredA steady signal gives no derivative contribution after transients settle. Rapid changes can produce large corrections.

Always check whether the controller is in automatic, operating against limits, or tracking a downstream device.

Proportional gain, proportional band and offset

Let e=SP−PVe=SP-PV with both values scaled to percent of input span. For a loop in which raising the command raises the PV, use the ideal proportional equation CO=b+KceCO=b+K_c e, with positive gain KcK_c and bias bb in percent output. The change for a change in error is ΔCO=KcΔe\Delta CO=K_c\Delta e until a limit applies. If the installed process needs opposite action, configure the sign according to its controller.

When both spans are normalized, proportional band is the percent of input span corresponding to a 100-point change in the unconstrained proportional output. PB=100%/∣Kc∣\text{PB}=100\%/|K_c|. A 40% band corresponds to gain magnitude 2.5. On a 0–500 psig transmitter, a 25 psig change represents 5 percentage points of input span. The output change then has magnitude 2.5(5)=12.52.5(5)=12.5 percentage points. A band of 200% represents gain magnitude 0.5. The installed device's parameter may instead be entered in engineering units. Check scaling before converting.

With a fixed bias, a P-only loop needs a nonzero error at equilibrium if the command required there differs from that bias. For example, suppose a heat exchanger's original output was 50% at zero error. After a load change, model the local steady relation as

PV−SP=CO−70PV-SP = CO-70

with both sides in their normalized percentage-point units. At the setpoint this plant would require 70% output. With Kc=2K_c=2 and bias 50%, the P-only controller gives CO=50−2(PV−SP)CO=50-2(PV-SP). Substituting yields 3(PV−SP)=−203(PV-SP)=-20. The resulting static equilibrium is 6.67 input-span points below setpoint, with output about 63.33%. This equilibrium assumes the loop settles and the output remains within limits. More gain can reduce this offset in the model while affecting dynamic stability on the actual loop.

One convention for integral and derivative

A continuous, dependent-gain model with derivative on error and an initial integral contribution included in the bias is:

CO(t)=b+Kc[e(t)+1Ti∫0te(ξ) dξ+Tdde(t)dt].CO(t)=b+K_c\left[e(t)+\frac{1}{T_i}\int_{0}^{t}e(\xi)\,d\xi+T_d\frac{de(t)}{dt}\right].

This ideal equation omits sampling, filtering and limits. In the dependent model, the outer gain affects all three terms. In the parallel model, the coefficients multiplying error, its integral and its derivative are separate. For the same error derivative and consistent time units, the parallel coefficients are KcK_c, Kc/TiK_c/T_i and KcTdK_cT_d. A series or interacting controller uses a different arrangement. Rockwell's Logix PID documentation gives an example of selectable independent and dependent forms with different configuration units.

In the dependent form, a constant error adds one proportional correction to the integral contribution in each reset time TiT_i. When Ti=0.25T_i=0.25 min per repeat, reset rate is 1/Ti=41/T_i=4 repeats per minute. In the parallel model with time measured in minutes and Kc=2K_c=2, the corresponding integral coefficient is 2/0.25=82/0.25=8 per minute. To enter this coefficient in a controller using seconds, use 8/60≈0.13338/60\approx0.1333 per second. These values describe the same ideal action. Changing only the proportional gain affects integral action in the dependent equation. A stable loop with integral action and sufficient control authority can remove steady offset after a constant disturbance. Saturation, an unreachable SP or deliberate integration limits can leave error.

Detect integration beyond available control

An output limit or a downstream selection can prevent the applied command from following the internal calculation. If the integral term keeps adding in the direction of the limit, the controller can remain saturated after the process recovers enough to need less output. A trend that shows the integral term accumulating with demanded output above the applied limit distinguishes this mechanism from a valve sticking at a lower command.

Monitor both the requested and applied output. Some controllers stop accumulation at configured output limits. Others condition integration or use back-calculation to track a limited or selected output. Where the actuator or inner loop imposes a separate limit, configured status or external reset feedback can make that downstream limit visible to the controller. For example, the Rockwell PlantPAx PPID documents both internal saturation handling and external windup inputs. Confirm mode tracking and restoration with the installed configuration.

Derivative on error or measured PV

At constant setpoint and e=SP−PVe=SP-PV, de/dt=−dPV/dtde/dt=-dPV/dt. In the dependent equation, derivative on measured PV uses −KcTd dPV/dt-K_cT_d\,dPV/dt. Derivative on error responds to the SP step as well as to PV motion. A derivative term acting on PV avoids the direct derivative kick from an SP step. The proportional term may still change the output at the same instant. Derivative estimates can amplify high-frequency measurement noise. Filtering and suitable timing limit that amplification. PI is often a practical choice for a noisy loop that needs integral correction. Evaluate any derivative use against the measured dynamics, noise and final-element response for that application.

Test Your Knowledge

A P-only pressure controller uses positive gain on e = SP − PV and has a normalized proportional band of 40%. The 0–500 psig transmitter rises from 250 to 275 psig with SP fixed. The output remains within its limits. What gain magnitude and signed output change result?

A

0.40 gain and a decrease of 2.0 output points

B

4.0 gain and an increase of 20.0 output points

C

2.5 gain and a decrease of 12.5 output points

D

2.5 gain and an increase of 25.0 output points

Test Your Knowledge

A warmup trend shows the steam valve following the applied output at its upper limit. The internal integral contribution keeps rising after the applied output reaches that limit. When PV reaches SP the calculated output still exceeds the applied limit. What mechanism is demonstrated?

A

Integral windup. Use anti-windup or tracking appropriate to the applied limit.

B

Derivative kick on SP. Use a higher derivative gain to leave the limit sooner.

C

Valve stiction. Recalibrate the travel feedback to reduce the integral contribution.

D

Normal proportional offset. Widen the band to make internal and applied outputs match.

Test Your Knowledge

A noisy flow signal is steady on average. A trend shows that the rapid output excursions coincide with the derivative contribution. The setpoint is fixed and the proportional and integral contributions are comparatively smooth. What does this evidence support?

A

Setpoint derivative kick is causing the excursions.

B

Derivative is amplifying measurement noise. Assess the PV signal, derivative filtering and whether PI meets the loop objective.

C

The integral contribution is winding up against a limit.

D

The valve is mechanically seized because the setpoint is fixed.

Sections you finish are checked off in the contents.